大型火灾图像数据集的设计及建立

Establish a large-scale fire image dataset

当前火灾图像识别方法主要依赖于大数据集,在样本不足时会出现探测结果不可靠的问题。针对国内外普遍缺乏火灾图像样本的问题,根据火灾探测算法开发和评估需求,建立样本数量充足且包含多种应用场景的大型火灾图像数据集。为确保火灾图像数据集完备与有效,通过数据集开发并评估多个图像型火灾探测算法,分别对数据集的开发有效性和评估有效性进行试验。试验结果表明:数据集具有较好的普适性和有效性,可以为图像型火灾探测算法的研究提供基础平台。

The current fire image recognition methods rely primarily on large datasets, which will cause unreliable detection results when training samples are insufficient. In response to the problem of generally lack of fire image samples, a largescale fire image dataset containing various application scenarios based on fire detection algorithm development and evaluation needs is proposed. This paper trains and develops fire detection algorithms based on the fire image dataset, and verifies the completeness and effectiveness of the fire image dataset. Experiments on the fire image dataset established in this paper show the universality and effectiveness of fire detection algorithms,the experimental results show that this fire image dataset provides a basic platform for the research of image-type fire detection algorithms.